Q100485: Information about Nuke's Multi-GPU support

Pete

Updated: February 03, 2026 00:20

Created: January 20, 2026 15:59

SUMMARY

This article will provide more information about Multi-GPU support in Nuke.

 

MORE INFORMATION

Our documentation specifies that Multi-GPU processing is available for Windows, macOS and Linux for identical GPUs in the same machine (for example two NVIDIA Quadro P4000s or two AMD Pro W7900s). However, Nuke will recognize whether you have more than two identical GPUs and will allow you to use all of them.

NOTE: Nuke is not currently tested against setups with more than two GPUs, so Multi-GPU processing in Nuke is supported for machines with two identical GPUs only.

When a machine has more than one GPU, such as two NVIDIA Quadro P4000 cards, the following information will be displayed when a GPU accelerated node like ZDefocus is created:

Although Nuke supports some Multi-GPU setups, only certain GPU accelerated nodes will benefit from having multiple GPUs, other GPU accelerated nodes will not benefit at all, and nodes which aren't GPU accelerated will see no difference.

For example, MotionBlur, Kronos, OFlow, VectorGenerator nodes can utilize up to 2 GPUs, while VectorBlur, SmartVector, VectorDisort and CaraVR nodes only use 1 GPU, and the ZDefocus and Convolve nodes should work with any number of GPUs.

One special case regarding Multi-GPU setups is the CopyCat node, which can utilize more than one GPU to train a single network, or run different training sessions on each individual GPU to train multiple models simultaneously. However, please note that these Multi-GPU performance boosts do not apply to Cattery models or other AIR nodes like Inference.

FURTHER READING

More information about Nuke’s system requirements can be found on the following page of our website: https://www.foundry.com/products/nuke/requirements

For more details regarding CopyCat and Multi-GPU devices, please visit the Training on Multiple GPUs section of our documentation.